Rethinking ME/CFS Diagnostic Reference Intervals via Machine Learning, and the Utility of Activin B for Defining Symptom Severity.
Lidbury, Brett A, Kita, Badia, Richardson, Alice M et al. · Diagnostics (Basel, Switzerland) · 2019 · DOI
Quick Summary
This study looked for blood test markers that could help diagnose ME/CFS and measure how severe someone's illness is. Researchers tested a protein called activin B along with routine blood work from pathology labs in people with ME/CFS and healthy controls. They used a computer learning method to see which combinations of blood markers worked best for identifying ME/CFS and predicting symptom severity.
Why It Matters
ME/CFS currently lacks objective diagnostic criteria, forcing reliance on symptom-based diagnosis. Identifying reliable blood biomarkers like activin B could enable faster, objective diagnosis and provide clinicians with tools to objectively measure disease severity—potentially improving patient outcomes and enabling better monitoring of disease progression.
Observed Findings
Serum activin B was significantly different between ME/CFS and control participants, though median levels were lower in the ME/CFS cohort than in the pilot study.
Five routine pathology markers predicted ME/CFS diagnosis with ≥62% accuracy using Random Forest modeling.
Addition of activin B to the pathology marker panel improved prediction specifically for mild-to-moderate ME/CFS cases.
24-h urinary creatinine clearance, serum urea, and serum activin B showed the strongest potential as diagnostic markers.
Activin B levels correlated with symptom severity as measured by weighted standing time (WST) classification.
Inferred Conclusions
Activin B combined with routine pathology markers may improve diagnostic accuracy for ME/CFS, particularly in mild-to-moderate disease.
New reference intervals for activin B and associated pathology markers could refine diagnostic approaches for ME/CFS.
Activin B has potential utility not only for diagnosis but also for objectively assessing symptom severity.
Machine learning approaches may help identify optimal combinations of routine blood markers for clinical diagnosis.
Remaining Questions
Why were activin B levels lower in the ME/CFS cohort in this study compared to the pilot study, and what accounts for this discrepancy?
What This Study Does Not Prove
This study does not prove that activin B causes ME/CFS or explain the biological mechanism behind elevation. The cross-sectional design cannot establish causality or temporal relationships. Results require validation in independent prospective cohorts before clinical implementation; machine learning models trained on one dataset may not generalize reliably to different populations.